def remove_services(client, node_type, logger):
"""
Remove all services managed by OVS
:param client: Client on which to remove the services
:type client: ovs.extensions.generic.sshclient.SSHClient
:param node_type: Type of node, can be 'master' or 'extra'
:type node_type: str
:param logger: Logger object used for logging
:type logger: ovs.log.log_handler.LogHandler
:return: None
"""
Toolbox.log(logger=logger, messages="Removing services")
stop_only = ["rabbitmq-server", "memcached"]
services = ["workers", "support-agent", "watcher-framework"]
if node_type == "master":
services += ["scheduled-tasks", "webapp-api", "volumerouter-consumer"]
if Toolbox.is_service_internally_managed(service="rabbitmq") is True:
services.append("rabbitmq-server")
if Toolbox.is_service_internally_managed(service="memcached") is True:
services.append("memcached")
for service in services:
if ServiceManager.has_service(service, client=client):
Toolbox.log(
logger=logger,
messages="{0} service {1}".format("Removing" if service not in stop_only else "Stopping", service),
)
ServiceManager.stop_service(service, client=client)
if service not in stop_only:
ServiceManager.remove_service(service, client=client)
def remove_services(client, node_type, logger):
"""
Remove all services managed by OVS
:param client: Client on which to remove the services
:type client: ovs_extensions.generic.sshclient.SSHClient
:param node_type: Type of node, can be 'master' or 'extra'
:type node_type: str
:param logger: Logger object used for logging
:type logger: ovs.extensions.generic.logger.Logger
:return: None
"""
Toolbox.log(logger=logger, messages='Removing services')
service_manager = ServiceFactory.get_manager()
stop_only = ['rabbitmq-server', 'memcached']
services = ['workers', 'support-agent', 'watcher-framework']
if node_type == 'master':
services += [
'scheduled-tasks', 'webapp-api', 'volumerouter-consumer'
]
if Toolbox.is_service_internally_managed(
service='rabbitmq') is True:
services.append('rabbitmq-server')
if Toolbox.is_service_internally_managed(
service='memcached') is True:
services.append('memcached')
for service in services:
if service_manager.has_service(service, client=client):
Toolbox.log(
logger=logger,
messages='{0} service {1}'.format(
'Removing' if service not in stop_only else 'Stopping',
service))
service_manager.stop_service(service, client=client)
if service not in stop_only:
service_manager.remove_service(service, client=client)
def remove_node(node_ip, silent=None):
"""
Remove the node with specified IP from the cluster
:param node_ip: IP of the node to remove
:type node_ip: str
:param silent: If silent == '--force-yes' no question will be asked to confirm the removal
:type silent: str
:return: None
"""
from ovs.dal.lists.storagerouterlist import StorageRouterList
from ovs.lib.storagedriver import StorageDriverController
from ovs.lib.vpool import VPoolController
Toolbox.log(logger=NodeRemovalController._logger,
messages='Remove node',
boxed=True)
Toolbox.log(
logger=NodeRemovalController._logger,
messages=
'WARNING: Some of these steps may take a very long time, please check the logs for more information\n\n'
)
service_manager = ServiceFactory.get_manager()
###############
# VALIDATIONS #
###############
try:
node_ip = node_ip.strip()
if not isinstance(node_ip, str):
raise ValueError('Node IP must be a string')
if not re.match(SSHClient.IP_REGEX, node_ip):
raise ValueError('Invalid IP {0} specified'.format(node_ip))
storage_router_all = sorted(StorageRouterList.get_storagerouters(),
key=lambda k: k.name)
storage_router_masters = StorageRouterList.get_masters()
storage_router_all_ips = set(
[storage_router.ip for storage_router in storage_router_all])
storage_router_master_ips = set([
storage_router.ip for storage_router in storage_router_masters
])
storage_router_to_remove = StorageRouterList.get_by_ip(node_ip)
offline_reasons = {}
if node_ip not in storage_router_all_ips:
raise ValueError(
'Unknown IP specified\nKnown in model:\n - {0}\nSpecified for removal:\n - {1}'
.format('\n - '.join(storage_router_all_ips), node_ip))
if len(storage_router_all_ips) == 1:
raise RuntimeError("Removing the only node is not possible")
if node_ip in storage_router_master_ips and len(
storage_router_master_ips) == 1:
raise RuntimeError(
"Removing the only master node is not possible")
if System.get_my_storagerouter() == storage_router_to_remove:
raise RuntimeError(
'The node to be removed cannot be identical to the node on which the removal is initiated'
)
Toolbox.log(
logger=NodeRemovalController._logger,
messages='Creating SSH connections to remaining master nodes')
master_ip = None
ip_client_map = {}
storage_routers_offline = []
storage_router_to_remove_online = True
for storage_router in storage_router_all:
try:
client = SSHClient(storage_router,
username='******',
timeout=10)
except (UnableToConnectException, NotAuthenticatedException,
TimeOutException) as ex:
if isinstance(ex, UnableToConnectException):
msg = 'Unable to connect'
elif isinstance(ex, NotAuthenticatedException):
msg = 'Could not authenticate'
elif isinstance(ex, TimeOutException):
msg = 'Connection timed out'
Toolbox.log(
logger=NodeRemovalController._logger,
messages=' * Node with IP {0:<15}- {1}'.format(
storage_router.ip, msg))
offline_reasons[storage_router.ip] = msg
storage_routers_offline.append(storage_router)
if storage_router == storage_router_to_remove:
storage_router_to_remove_online = False
continue
Toolbox.log(
logger=NodeRemovalController._logger,
messages=' * Node with IP {0:<15}- Successfully connected'
.format(storage_router.ip))
ip_client_map[storage_router.ip] = client
if storage_router != storage_router_to_remove and storage_router.node_type == 'MASTER':
master_ip = storage_router.ip
if len(ip_client_map) == 0 or master_ip is None:
raise RuntimeError(
'Could not connect to any master node in the cluster')
storage_router_to_remove.invalidate_dynamics('vdisks_guids')
if len(
storage_router_to_remove.vdisks_guids
) > 0: # vDisks are supposed to be moved away manually before removing a node
raise RuntimeError(
"Still vDisks attached to Storage Router {0}".format(
storage_router_to_remove.name))
internal_memcached = Toolbox.is_service_internally_managed(
service='memcached')
internal_rabbit_mq = Toolbox.is_service_internally_managed(
service='rabbitmq')
memcached_endpoints = Configuration.get(
key='/ovs/framework/memcache|endpoints')
rabbit_mq_endpoints = Configuration.get(
key='/ovs/framework/messagequeue|endpoints')
copy_memcached_endpoints = list(memcached_endpoints)
copy_rabbit_mq_endpoints = list(rabbit_mq_endpoints)
for endpoint in memcached_endpoints:
if endpoint.startswith(storage_router_to_remove.ip):
copy_memcached_endpoints.remove(endpoint)
for endpoint in rabbit_mq_endpoints:
if endpoint.startswith(storage_router_to_remove.ip):
copy_rabbit_mq_endpoints.remove(endpoint)
if len(copy_memcached_endpoints
) == 0 and internal_memcached is True:
raise RuntimeError(
'Removal of provided nodes will result in a complete removal of the memcached service'
)
if len(copy_rabbit_mq_endpoints
) == 0 and internal_rabbit_mq is True:
raise RuntimeError(
'Removal of provided nodes will result in a complete removal of the messagequeue service'
)
Toolbox.run_hooks(component='noderemoval',
sub_component='validate_removal',
logger=NodeRemovalController._logger,
cluster_ip=storage_router_to_remove.ip)
except KeyboardInterrupt:
Toolbox.log(logger=NodeRemovalController._logger, messages='\n')
Toolbox.log(
logger=NodeRemovalController._logger,
messages=
'Removal has been aborted during the validation step. No changes have been applied.',
boxed=True,
loglevel='warning')
sys.exit(1)
except Exception as exception:
Toolbox.log(logger=NodeRemovalController._logger,
messages=[str(exception)],
boxed=True,
loglevel='exception')
sys.exit(1)
#################
# CONFIRMATIONS #
#################
try:
interactive = silent != '--force-yes'
remove_asd_manager = not interactive # Remove ASD manager if non-interactive else ask
if interactive is True:
if len(storage_routers_offline) > 0:
Toolbox.log(
logger=NodeRemovalController._logger,
messages=
'Certain nodes appear to be offline. These will not fully removed and will cause issues if they are not really offline.'
)
Toolbox.log(
logger=NodeRemovalController._logger,
messages='Offline nodes: {0}'.format(''.join(
('\n * {0:<15}- {1}.'.format(ip, message)
for ip, message in offline_reasons.iteritems()))))
valid_node_info = Interactive.ask_yesno(
message=
'Continue the removal with these being presumably offline?',
default_value=False)
if valid_node_info is False:
Toolbox.log(
logger=NodeRemovalController._logger,
messages=
'Please validate the state of the nodes before removing.',
title=True)
sys.exit(1)
proceed = Interactive.ask_yesno(
message='Are you sure you want to remove node {0}?'.format(
storage_router_to_remove.name),
default_value=False)
if proceed is False:
Toolbox.log(logger=NodeRemovalController._logger,
messages='Abort removal',
title=True)
sys.exit(1)
remove_asd_manager = True
if storage_router_to_remove_online is True:
client = SSHClient(endpoint=storage_router_to_remove,
username='******')
if service_manager.has_service(name='asd-manager',
client=client):
remove_asd_manager = Interactive.ask_yesno(
message=
'Do you also want to remove the ASD manager and related ASDs?',
default_value=False)
if remove_asd_manager is True or storage_router_to_remove_online is False:
for fct in Toolbox.fetch_hooks('noderemoval',
'validate_asd_removal'):
validation_output = fct(storage_router_to_remove.ip)
if validation_output['confirm'] is True:
if Interactive.ask_yesno(
message=validation_output['question'],
default_value=False) is False:
remove_asd_manager = False
break
except KeyboardInterrupt:
Toolbox.log(logger=NodeRemovalController._logger, messages='\n')
Toolbox.log(
logger=NodeRemovalController._logger,
messages=
'Removal has been aborted during the confirmation step. No changes have been applied.',
boxed=True,
loglevel='warning')
sys.exit(1)
except Exception as exception:
Toolbox.log(logger=NodeRemovalController._logger,
messages=[str(exception)],
boxed=True,
loglevel='exception')
sys.exit(1)
###########
# REMOVAL #
###########
try:
Toolbox.log(logger=NodeRemovalController._logger,
messages='Starting removal of node {0} - {1}'.format(
storage_router_to_remove.name,
storage_router_to_remove.ip))
if storage_router_to_remove_online is False:
Toolbox.log(
logger=NodeRemovalController._logger,
messages=
' Marking all Storage Drivers served by Storage Router {0} as offline'
.format(storage_router_to_remove.ip))
StorageDriverController.mark_offline(
storagerouter_guid=storage_router_to_remove.guid)
# Remove vPools
Toolbox.log(logger=NodeRemovalController._logger,
messages=' Removing vPools from node'.format(
storage_router_to_remove.ip))
storage_routers_offline_guids = [
sr.guid for sr in storage_routers_offline
if sr.guid != storage_router_to_remove.guid
]
for storage_driver in storage_router_to_remove.storagedrivers:
Toolbox.log(logger=NodeRemovalController._logger,
messages=' Removing vPool {0} from node'.format(
storage_driver.vpool.name))
VPoolController.shrink_vpool(
storagedriver_guid=storage_driver.guid,
offline_storage_router_guids=storage_routers_offline_guids)
# Demote if MASTER
if storage_router_to_remove.node_type == 'MASTER':
NodeTypeController.demote_node(
cluster_ip=storage_router_to_remove.ip,
master_ip=master_ip,
ip_client_map=ip_client_map,
unique_id=storage_router_to_remove.machine_id,
unconfigure_memcached=internal_memcached,
unconfigure_rabbitmq=internal_rabbit_mq,
offline_nodes=storage_routers_offline)
# Stop / remove services
Toolbox.log(logger=NodeRemovalController._logger,
messages='Stopping and removing services')
if storage_router_to_remove_online is True:
client = SSHClient(endpoint=storage_router_to_remove,
username='******')
NodeRemovalController.remove_services(
client=client,
node_type=storage_router_to_remove.node_type.lower(),
logger=NodeRemovalController._logger)
service = 'watcher-config'
if service_manager.has_service(service, client=client):
Toolbox.log(
logger=NodeRemovalController._logger,
messages='Removing service {0}'.format(service))
service_manager.stop_service(service, client=client)
service_manager.remove_service(service, client=client)
Toolbox.run_hooks(component='noderemoval',
sub_component='remove',
logger=NodeRemovalController._logger,
cluster_ip=storage_router_to_remove.ip,
complete_removal=remove_asd_manager)
# Clean up model
Toolbox.log(logger=NodeRemovalController._logger,
messages='Removing node from model')
for service in storage_router_to_remove.services:
service.delete()
for disk in storage_router_to_remove.disks:
for partition in disk.partitions:
partition.delete()
disk.delete()
for j_domain in storage_router_to_remove.domains:
j_domain.delete()
Configuration.delete('/ovs/framework/hosts/{0}'.format(
storage_router_to_remove.machine_id))
NodeTypeController.restart_framework_and_memcache_services(
clients=ip_client_map,
offline_node_ips=[node.ip for node in storage_routers_offline],
logger=NodeRemovalController._logger)
if storage_router_to_remove_online is True:
client = SSHClient(endpoint=storage_router_to_remove,
username='******')
client.file_delete(filenames=[CACC_LOCATION])
client.file_delete(filenames=[CONFIG_STORE_LOCATION])
storage_router_to_remove.delete()
Toolbox.log(logger=NodeRemovalController._logger,
messages='Successfully removed node\n')
except Exception as exception:
Toolbox.log(logger=NodeRemovalController._logger, messages='\n')
Toolbox.log(
logger=NodeRemovalController._logger,
messages=['An unexpected error occurred:',
str(exception)],
boxed=True,
loglevel='exception')
sys.exit(1)
except KeyboardInterrupt:
Toolbox.log(logger=NodeRemovalController._logger, messages='\n')
Toolbox.log(
logger=NodeRemovalController._logger,
messages=
'This setup was aborted. Open vStorage may be in an inconsistent state, make sure to validate the installation.',
boxed=True,
loglevel='error')
sys.exit(1)
if remove_asd_manager is True and storage_router_to_remove_online is True:
Toolbox.log(logger=NodeRemovalController._logger,
messages='\nRemoving ASD Manager')
with remote(storage_router_to_remove.ip, [os]) as rem:
rem.os.system('asd-manager remove --force-yes')
Toolbox.log(logger=NodeRemovalController._logger,
messages='Remove nodes finished',
title=True)
def remove_node(node_ip, silent=None):
"""
Remove the node with specified IP from the cluster
:param node_ip: IP of the node to remove
:type node_ip: str
:param silent: If silent == '--force-yes' no question will be asked to confirm the removal
:type silent: str
:return: None
"""
from ovs.lib.storagedriver import StorageDriverController
from ovs.lib.storagerouter import StorageRouterController
from ovs.dal.lists.storagerouterlist import StorageRouterList
Toolbox.log(logger=NodeRemovalController._logger, messages="Remove node", boxed=True)
Toolbox.log(
logger=NodeRemovalController._logger,
messages="WARNING: Some of these steps may take a very long time, please check the logs for more information\n\n",
)
###############
# VALIDATIONS #
###############
try:
node_ip = node_ip.strip()
if not isinstance(node_ip, str):
raise ValueError("Node IP must be a string")
if not re.match(SSHClient.IP_REGEX, node_ip):
raise ValueError("Invalid IP {0} specified".format(node_ip))
storage_router_all = StorageRouterList.get_storagerouters()
storage_router_masters = StorageRouterList.get_masters()
storage_router_all_ips = set([storage_router.ip for storage_router in storage_router_all])
storage_router_master_ips = set([storage_router.ip for storage_router in storage_router_masters])
storage_router_to_remove = StorageRouterList.get_by_ip(node_ip)
if node_ip not in storage_router_all_ips:
raise ValueError(
"Unknown IP specified\nKnown in model:\n - {0}\nSpecified for removal:\n - {1}".format(
"\n - ".join(storage_router_all_ips), node_ip
)
)
if len(storage_router_all_ips) == 1:
raise RuntimeError("Removing the only node is not possible")
if node_ip in storage_router_master_ips and len(storage_router_master_ips) == 1:
raise RuntimeError("Removing the only master node is not possible")
if System.get_my_storagerouter() == storage_router_to_remove:
raise RuntimeError(
"The node to be removed cannot be identical to the node on which the removal is initiated"
)
Toolbox.log(
logger=NodeRemovalController._logger, messages="Creating SSH connections to remaining master nodes"
)
master_ip = None
ip_client_map = {}
storage_routers_offline = []
storage_router_to_remove_online = True
for storage_router in storage_router_all:
try:
client = SSHClient(storage_router, username="******")
if client.run(["pwd"]):
Toolbox.log(
logger=NodeRemovalController._logger,
messages=" Node with IP {0:<15} successfully connected to".format(storage_router.ip),
)
ip_client_map[storage_router.ip] = client
if storage_router != storage_router_to_remove and storage_router.node_type == "MASTER":
master_ip = storage_router.ip
except UnableToConnectException:
Toolbox.log(
logger=NodeRemovalController._logger,
messages=" Node with IP {0:<15} is unreachable".format(storage_router.ip),
)
storage_routers_offline.append(storage_router)
if storage_router == storage_router_to_remove:
storage_router_to_remove_online = False
if len(ip_client_map) == 0 or master_ip is None:
raise RuntimeError("Could not connect to any master node in the cluster")
storage_router_to_remove.invalidate_dynamics("vdisks_guids")
if (
len(storage_router_to_remove.vdisks_guids) > 0
): # vDisks are supposed to be moved away manually before removing a node
raise RuntimeError("Still vDisks attached to Storage Router {0}".format(storage_router_to_remove.name))
internal_memcached = Toolbox.is_service_internally_managed(service="memcached")
internal_rabbit_mq = Toolbox.is_service_internally_managed(service="rabbitmq")
memcached_endpoints = Configuration.get(key="/ovs/framework/memcache|endpoints")
rabbit_mq_endpoints = Configuration.get(key="/ovs/framework/messagequeue|endpoints")
copy_memcached_endpoints = list(memcached_endpoints)
copy_rabbit_mq_endpoints = list(rabbit_mq_endpoints)
for endpoint in memcached_endpoints:
if endpoint.startswith(storage_router_to_remove.ip):
copy_memcached_endpoints.remove(endpoint)
for endpoint in rabbit_mq_endpoints:
if endpoint.startswith(storage_router_to_remove.ip):
copy_rabbit_mq_endpoints.remove(endpoint)
if len(copy_memcached_endpoints) == 0 and internal_memcached is True:
raise RuntimeError(
"Removal of provided nodes will result in a complete removal of the memcached service"
)
if len(copy_rabbit_mq_endpoints) == 0 and internal_rabbit_mq is True:
raise RuntimeError(
"Removal of provided nodes will result in a complete removal of the messagequeue service"
)
except Exception as exception:
Toolbox.log(
logger=NodeRemovalController._logger, messages=[str(exception)], boxed=True, loglevel="exception"
)
sys.exit(1)
#################
# CONFIRMATIONS #
#################
interactive = silent != "--force-yes"
remove_asd_manager = not interactive # Remove ASD manager if non-interactive else ask
if interactive is True:
proceed = Interactive.ask_yesno(
message="Are you sure you want to remove node {0}?".format(storage_router_to_remove.name),
default_value=False,
)
if proceed is False:
Toolbox.log(logger=NodeRemovalController._logger, messages="Abort removal", title=True)
sys.exit(1)
if storage_router_to_remove_online is True:
client = SSHClient(endpoint=storage_router_to_remove, username="******")
if ServiceManager.has_service(name="asd-manager", client=client):
remove_asd_manager = Interactive.ask_yesno(
message="Do you also want to remove the ASD manager and related ASDs?", default_value=False
)
if remove_asd_manager is True or storage_router_to_remove_online is False:
for function in Toolbox.fetch_hooks("setup", "validate_asd_removal"):
validation_output = function(storage_router_to_remove.ip)
if validation_output["confirm"] is True:
if Interactive.ask_yesno(message=validation_output["question"], default_value=False) is False:
remove_asd_manager = False
break
###########
# REMOVAL #
###########
try:
Toolbox.log(
logger=NodeRemovalController._logger,
messages="Starting removal of node {0} - {1}".format(
storage_router_to_remove.name, storage_router_to_remove.ip
),
)
if storage_router_to_remove_online is False:
Toolbox.log(
logger=NodeRemovalController._logger,
messages=" Marking all Storage Drivers served by Storage Router {0} as offline".format(
storage_router_to_remove.ip
),
)
StorageDriverController.mark_offline(storagerouter_guid=storage_router_to_remove.guid)
# Remove vPools
Toolbox.log(
logger=NodeRemovalController._logger,
messages=" Removing vPools from node".format(storage_router_to_remove.ip),
)
storage_routers_offline_guids = [
sr.guid for sr in storage_routers_offline if sr.guid != storage_router_to_remove.guid
]
for storage_driver in storage_router_to_remove.storagedrivers:
Toolbox.log(
logger=NodeRemovalController._logger,
messages=" Removing vPool {0} from node".format(storage_driver.vpool.name),
)
StorageRouterController.remove_storagedriver(
storagedriver_guid=storage_driver.guid, offline_storage_router_guids=storage_routers_offline_guids
)
# Demote if MASTER
if storage_router_to_remove.node_type == "MASTER":
NodeTypeController.demote_node(
cluster_ip=storage_router_to_remove.ip,
master_ip=master_ip,
ip_client_map=ip_client_map,
unique_id=storage_router_to_remove.machine_id,
unconfigure_memcached=internal_memcached,
unconfigure_rabbitmq=internal_rabbit_mq,
offline_nodes=storage_routers_offline,
)
# Stop / remove services
Toolbox.log(logger=NodeRemovalController._logger, messages="Stopping and removing services")
config_store = Configuration.get_store()
if storage_router_to_remove_online is True:
client = SSHClient(endpoint=storage_router_to_remove, username="******")
NodeRemovalController.remove_services(
client=client,
node_type=storage_router_to_remove.node_type.lower(),
logger=NodeRemovalController._logger,
)
service = "watcher-config"
if ServiceManager.has_service(service, client=client):
Toolbox.log(logger=NodeRemovalController._logger, messages="Removing service {0}".format(service))
ServiceManager.stop_service(service, client=client)
ServiceManager.remove_service(service, client=client)
if config_store == "etcd":
from ovs.extensions.db.etcd.installer import EtcdInstaller
if Configuration.get(key="/ovs/framework/external_config") is None:
Toolbox.log(logger=NodeRemovalController._logger, messages=" Removing Etcd cluster")
try:
EtcdInstaller.stop("config", client)
EtcdInstaller.remove("config", client)
except Exception as ex:
Toolbox.log(
logger=NodeRemovalController._logger,
messages=["\nFailed to unconfigure Etcd", ex],
loglevel="exception",
)
Toolbox.log(logger=NodeRemovalController._logger, messages="Removing Etcd proxy")
EtcdInstaller.remove_proxy("config", client.ip)
Toolbox.run_hooks(
component="noderemoval",
sub_component="remove",
logger=NodeRemovalController._logger,
cluster_ip=storage_router_to_remove.ip,
complete_removal=remove_asd_manager,
)
# Clean up model
Toolbox.log(logger=NodeRemovalController._logger, messages="Removing node from model")
for service in storage_router_to_remove.services:
service.delete()
for disk in storage_router_to_remove.disks:
for partition in disk.partitions:
partition.delete()
disk.delete()
for j_domain in storage_router_to_remove.domains:
j_domain.delete()
Configuration.delete("/ovs/framework/hosts/{0}".format(storage_router_to_remove.machine_id))
NodeTypeController.restart_framework_and_memcache_services(
clients=ip_client_map,
offline_node_ips=[node.ip for node in storage_routers_offline],
logger=NodeRemovalController._logger,
)
if storage_router_to_remove_online is True:
client = SSHClient(endpoint=storage_router_to_remove, username="******")
if config_store == "arakoon":
client.file_delete(filenames=[ArakoonConfiguration.CACC_LOCATION])
client.file_delete(filenames=[Configuration.BOOTSTRAP_CONFIG_LOCATION])
storage_router_to_remove.delete()
Toolbox.log(logger=NodeRemovalController._logger, messages="Successfully removed node\n")
except Exception as exception:
Toolbox.log(logger=NodeRemovalController._logger, messages="\n")
Toolbox.log(
logger=NodeRemovalController._logger,
messages=["An unexpected error occurred:", str(exception)],
boxed=True,
loglevel="exception",
)
sys.exit(1)
except KeyboardInterrupt:
Toolbox.log(logger=NodeRemovalController._logger, messages="\n")
Toolbox.log(
logger=NodeRemovalController._logger,
messages="This setup was aborted. Open vStorage may be in an inconsistent state, make sure to validate the installation.",
boxed=True,
loglevel="error",
)
sys.exit(1)
if remove_asd_manager is True:
Toolbox.log(logger=NodeRemovalController._logger, messages="\nRemoving ASD Manager")
with remote(storage_router_to_remove.ip, [os]) as rem:
rem.os.system("asd-manager remove --force-yes")
Toolbox.log(logger=NodeRemovalController._logger, messages="Remove nodes finished", title=True)
def promote_or_demote_node(node_action,
cluster_ip=None,
execute_rollback=False):
"""
Promotes or demotes the local node
:param node_action: Demote or promote
:type node_action: str
:param cluster_ip: IP of node to promote or demote
:type cluster_ip: str
:param execute_rollback: In case of failure revert the changes made
:type execute_rollback: bool
:return: None
"""
if node_action not in ('promote', 'demote'):
raise ValueError('Nodes can only be promoted or demoted')
Toolbox.log(logger=NodeTypeController._logger,
messages='Open vStorage Setup - {0}'.format(
node_action.capitalize()),
boxed=True)
try:
Toolbox.log(logger=NodeTypeController._logger,
messages='Collecting information',
title=True)
machine_id = System.get_my_machine_id()
if Configuration.get('/ovs/framework/hosts/{0}/setupcompleted'.
format(machine_id)) is False:
raise RuntimeError('No local OVS setup found.')
if cluster_ip and not re.match(Toolbox.regex_ip, cluster_ip):
raise RuntimeError(
'Incorrect IP provided ({0})'.format(cluster_ip))
if cluster_ip:
client = SSHClient(endpoint=cluster_ip)
machine_id = System.get_my_machine_id(client)
node_type = Configuration.get(
'/ovs/framework/hosts/{0}/type'.format(machine_id))
if node_action == 'promote' and node_type == 'MASTER':
raise RuntimeError('This node is already master.')
elif node_action == 'demote' and node_type == 'EXTRA':
raise RuntimeError('This node should be a master.')
elif node_type not in ['MASTER', 'EXTRA']:
raise RuntimeError('This node is not correctly configured.')
master_ip = None
offline_nodes = []
online = True
target_client = None
if node_action == 'demote' and cluster_ip: # Demote an offline node
from ovs.dal.lists.storagerouterlist import StorageRouterList
from ovs.lib.storagedriver import StorageDriverController
ip = cluster_ip
unique_id = None
ip_client_map = {}
for storage_router in StorageRouterList.get_storagerouters():
try:
client = SSHClient(storage_router.ip, username='******')
if storage_router.node_type == 'MASTER':
master_ip = storage_router.ip
ip_client_map[storage_router.ip] = client
except UnableToConnectException:
if storage_router.ip == cluster_ip:
online = False
unique_id = storage_router.machine_id
StorageDriverController.mark_offline(
storagerouter_guid=storage_router.guid)
offline_nodes.append(storage_router)
if online is True:
raise RuntimeError(
"If the node is online, please use 'ovs setup demote' executed on the node you wish to demote"
)
if master_ip is None:
raise RuntimeError(
'Failed to retrieve another responsive MASTER node')
else:
target_password = Toolbox.ask_validate_password(
ip='127.0.0.1', logger=NodeTypeController._logger)
target_client = SSHClient('127.0.0.1',
username='******',
password=target_password)
unique_id = System.get_my_machine_id(target_client)
ip = Configuration.get(
'/ovs/framework/hosts/{0}/ip'.format(unique_id))
storagerouter_info = NodeTypeController.retrieve_storagerouter_info_via_host(
ip=target_client.ip, password=target_password)
node_ips = [
sr_info['ip']
for sr_info in storagerouter_info.itervalues()
]
master_node_ips = [
sr_info['ip']
for sr_info in storagerouter_info.itervalues()
if sr_info['type'] == 'master' and sr_info['ip'] != ip
]
if len(master_node_ips) == 0:
if node_action == 'promote':
raise RuntimeError('No master node could be found')
else:
raise RuntimeError(
'It is not possible to remove the only master')
master_ip = master_node_ips[0]
ip_client_map = dict(
(node_ip, SSHClient(node_ip, username='******'))
for node_ip in node_ips)
if node_action == 'demote':
for cluster_name in Configuration.list('/ovs/arakoon'):
config = ArakoonClusterConfig(cluster_id=cluster_name)
arakoon_client = ArakoonInstaller.build_client(config)
metadata = json.loads(
arakoon_client.get(ArakoonInstaller.METADATA_KEY))
if len(config.nodes) == 1 and config.nodes[
0].ip == ip and metadata.get('internal') is True:
raise RuntimeError(
'Demote is not supported when single node Arakoon cluster(s) are present on the node to be demoted.'
)
configure_rabbitmq = Toolbox.is_service_internally_managed(
service='rabbitmq')
configure_memcached = Toolbox.is_service_internally_managed(
service='memcached')
if node_action == 'promote':
try:
NodeTypeController.promote_node(
cluster_ip=ip,
master_ip=master_ip,
ip_client_map=ip_client_map,
unique_id=unique_id,
configure_memcached=configure_memcached,
configure_rabbitmq=configure_rabbitmq)
except Exception:
if execute_rollback is True:
NodeTypeController.demote_node(
cluster_ip=ip,
master_ip=master_ip,
ip_client_map=ip_client_map,
unique_id=unique_id,
unconfigure_memcached=configure_memcached,
unconfigure_rabbitmq=configure_rabbitmq,
offline_nodes=offline_nodes)
elif target_client is not None:
target_client.file_write('/tmp/ovs_rollback', 'demote')
raise
else:
try:
NodeTypeController.demote_node(
cluster_ip=ip,
master_ip=master_ip,
ip_client_map=ip_client_map,
unique_id=unique_id,
unconfigure_memcached=configure_memcached,
unconfigure_rabbitmq=configure_rabbitmq,
offline_nodes=offline_nodes)
except Exception:
if execute_rollback is True:
NodeTypeController.promote_node(
cluster_ip=ip,
master_ip=master_ip,
ip_client_map=ip_client_map,
unique_id=unique_id,
configure_memcached=configure_memcached,
configure_rabbitmq=configure_rabbitmq)
elif target_client is not None:
target_client.file_write('/tmp/ovs_rollback',
'promote')
raise
Toolbox.log(logger=NodeTypeController._logger, messages='\n')
Toolbox.log(logger=NodeTypeController._logger,
messages='{0} complete.'.format(
node_action.capitalize()),
boxed=True)
except Exception as exception:
Toolbox.log(logger=NodeTypeController._logger, messages='\n')
Toolbox.log(
logger=NodeTypeController._logger,
messages=['An unexpected error occurred:',
str(exception)],
boxed=True,
loglevel='exception')
sys.exit(1)
except KeyboardInterrupt:
Toolbox.log(logger=NodeTypeController._logger, messages='\n')
Toolbox.log(
logger=NodeTypeController._logger,
messages=
'This setup was aborted. Open vStorage may be in an inconsistent state, make sure to validate the installation.',
boxed=True,
loglevel='error')
sys.exit(1)
def promote_or_demote_node(node_action, cluster_ip=None, execute_rollback=False):
"""
Promotes or demotes the local node
:param node_action: Demote or promote
:type node_action: str
:param cluster_ip: IP of node to promote or demote
:type cluster_ip: str
:param execute_rollback: In case of failure revert the changes made
:type execute_rollback: bool
:return: None
"""
if node_action not in ('promote', 'demote'):
raise ValueError('Nodes can only be promoted or demoted')
Toolbox.log(logger=NodeTypeController._logger, messages='Open vStorage Setup - {0}'.format(node_action.capitalize()), boxed=True)
try:
Toolbox.log(logger=NodeTypeController._logger, messages='Collecting information', title=True)
machine_id = System.get_my_machine_id()
if Configuration.get('/ovs/framework/hosts/{0}/setupcompleted'.format(machine_id)) is False:
raise RuntimeError('No local OVS setup found.')
if cluster_ip and not re.match(Toolbox.regex_ip, cluster_ip):
raise RuntimeError('Incorrect IP provided ({0})'.format(cluster_ip))
if cluster_ip:
client = SSHClient(endpoint=cluster_ip)
machine_id = System.get_my_machine_id(client)
node_type = Configuration.get('/ovs/framework/hosts/{0}/type'.format(machine_id))
if node_action == 'promote' and node_type == 'MASTER':
raise RuntimeError('This node is already master.')
elif node_action == 'demote' and node_type == 'EXTRA':
raise RuntimeError('This node should be a master.')
elif node_type not in ['MASTER', 'EXTRA']:
raise RuntimeError('This node is not correctly configured.')
master_ip = None
offline_nodes = []
online = True
target_client = None
if node_action == 'demote' and cluster_ip: # Demote an offline node
from ovs.dal.lists.storagerouterlist import StorageRouterList
from ovs.lib.storagedriver import StorageDriverController
ip = cluster_ip
unique_id = None
ip_client_map = {}
for storage_router in StorageRouterList.get_storagerouters():
try:
client = SSHClient(storage_router.ip, username='******')
if storage_router.node_type == 'MASTER':
master_ip = storage_router.ip
ip_client_map[storage_router.ip] = client
except UnableToConnectException:
if storage_router.ip == cluster_ip:
online = False
unique_id = storage_router.machine_id
StorageDriverController.mark_offline(storagerouter_guid=storage_router.guid)
offline_nodes.append(storage_router)
if online is True:
raise RuntimeError("If the node is online, please use 'ovs setup demote' executed on the node you wish to demote")
if master_ip is None:
raise RuntimeError('Failed to retrieve another responsive MASTER node')
else:
target_password = Toolbox.ask_validate_password(ip='127.0.0.1', logger=NodeTypeController._logger)
target_client = SSHClient('127.0.0.1', username='******', password=target_password)
unique_id = System.get_my_machine_id(target_client)
ip = Configuration.get('/ovs/framework/hosts/{0}/ip'.format(unique_id))
storagerouter_info = NodeTypeController.retrieve_storagerouter_info_via_host(ip=target_client.ip, password=target_password)
node_ips = [sr_info['ip'] for sr_info in storagerouter_info.itervalues()]
master_node_ips = [sr_info['ip'] for sr_info in storagerouter_info.itervalues() if sr_info['type'] == 'master' and sr_info['ip'] != ip]
if len(master_node_ips) == 0:
if node_action == 'promote':
raise RuntimeError('No master node could be found')
else:
raise RuntimeError('It is not possible to remove the only master')
master_ip = master_node_ips[0]
ip_client_map = dict((node_ip, SSHClient(node_ip, username='******')) for node_ip in node_ips)
if node_action == 'demote':
for cluster_name in Configuration.list('/ovs/arakoon'):
config = ArakoonClusterConfig(cluster_name, False)
config.load_config()
arakoon_client = ArakoonInstaller.build_client(config)
metadata = json.loads(arakoon_client.get(ArakoonInstaller.METADATA_KEY))
if len(config.nodes) == 1 and config.nodes[0].ip == ip and metadata.get('internal') is True:
raise RuntimeError('Demote is not supported when single node Arakoon cluster(s) are present on the node to be demoted.')
configure_rabbitmq = Toolbox.is_service_internally_managed(service='rabbitmq')
configure_memcached = Toolbox.is_service_internally_managed(service='memcached')
if node_action == 'promote':
try:
NodeTypeController.promote_node(cluster_ip=ip,
master_ip=master_ip,
ip_client_map=ip_client_map,
unique_id=unique_id,
configure_memcached=configure_memcached,
configure_rabbitmq=configure_rabbitmq)
except Exception:
if execute_rollback is True:
NodeTypeController.demote_node(cluster_ip=ip,
master_ip=master_ip,
ip_client_map=ip_client_map,
unique_id=unique_id,
unconfigure_memcached=configure_memcached,
unconfigure_rabbitmq=configure_rabbitmq,
offline_nodes=offline_nodes)
elif target_client is not None:
target_client.file_write('/tmp/ovs_rollback', 'demote')
raise
else:
try:
NodeTypeController.demote_node(cluster_ip=ip,
master_ip=master_ip,
ip_client_map=ip_client_map,
unique_id=unique_id,
unconfigure_memcached=configure_memcached,
unconfigure_rabbitmq=configure_rabbitmq,
offline_nodes=offline_nodes)
except Exception:
if execute_rollback is True:
NodeTypeController.promote_node(cluster_ip=ip,
master_ip=master_ip,
ip_client_map=ip_client_map,
unique_id=unique_id,
configure_memcached=configure_memcached,
configure_rabbitmq=configure_rabbitmq)
elif target_client is not None:
target_client.file_write('/tmp/ovs_rollback', 'promote')
raise
Toolbox.log(logger=NodeTypeController._logger, messages='\n')
Toolbox.log(logger=NodeTypeController._logger, messages='{0} complete.'.format(node_action.capitalize()), boxed=True)
except Exception as exception:
Toolbox.log(logger=NodeTypeController._logger, messages='\n')
Toolbox.log(logger=NodeTypeController._logger, messages=['An unexpected error occurred:', str(exception)], boxed=True, loglevel='exception')
sys.exit(1)
except KeyboardInterrupt:
Toolbox.log(logger=NodeTypeController._logger, messages='\n')
Toolbox.log(logger=NodeTypeController._logger,
messages='This setup was aborted. Open vStorage may be in an inconsistent state, make sure to validate the installation.',
boxed=True,
loglevel='error')
sys.exit(1)
